Companies are pouring billions into AI. It has yet to pay off.

Corporate spending on artificial intelligence is surging as executives bank on major efficiency gains. So far, they report little effect to the bottom line. (Antonio Sortino/The New York Times)

By Steve Lohr

Nearly four decades ago, when the personal computer boom was in full swing, a phenomenon known as the “productivity paradox” emerged.

It was a reference to how, despite companies’ huge investments in new technology, there was scant evidence of a corresponding gain in workers’ efficiency.

Today, the same paradox is appearing, but with generative artificial intelligence. According to recent research from McKinsey & Co., nearly 8 in 10 companies have reported using generative AI, but just as many have reported “no significant bottom-line impact.”

AI technology has been racing ahead with chatbots such as ChatGPT, fueled by a high-stakes arms race among tech giants and superrich startups and prompting an expectation that everything from back-office accounting to customer service will be revolutionized. But the payoff for businesses outside the tech sector is lagging behind, plagued by issues including an irritating tendency by chatbots to make stuff up.

That means that businesses will have to continue to invest billions to avoid falling behind — but it could be years before the technology delivers an economywide payoff, as companies gradually figure out what works best.

Call it the “the gen. AI paradox,” as McKinsey did in its research report. Investments in generative AI by businesses are expected to increase 94% this year to $61.9 billion, according to IDC, a technology research firm.

But the percentage of companies abandoning most of their AI pilot projects soared to 42% by the end of 2024, up from 17% the previous year, according to a survey of more than 1,000 technology and business managers by S&P Global, a data and analytics firm.

Projects failed not only because of technical hurdles, but often because of “human factors” like employee and customer resistance or lack of skills, said Alexander Johnston, a senior analyst at S&P Global.

Gartner, a research and advisory firm that charts technological “hype cycles,” predicts that AI is sliding toward a stage it calls “the trough of disillusionment.” The low point is expected next year, before the technology eventually becomes a proven productivity tool, said John-David Lovelock, the chief forecaster at Gartner.

That was the pattern with past technologies such as personal computers and the internet — early exuberance, the hard slog of mastering a technology, followed by a transformation of industries and work.

The winners so far have been the suppliers of AI technology and advice. They include Microsoft, Amazon and Google, which offer AI software, while Nvidia is the runaway leader in AI chips. Executives at those companies have bragged how AI is reshaping their own workforces, eliminating the need for some entry-level coding work and making other workers more efficient.

AI will eventually replace entire swaths of human employees, many predict, a perspective that is being widely embraced and echoed in the corporate mainstream. At the Aspen Ideas Festival in June, Jim Farley, the CEO of Ford Motor Co., said, “Artificial intelligence is going to replace literally half of all white-collar workers in the U.S.”

Whether that type of revolutionary change occurs, and how soon, depends on the real-world testing ground of many businesses.

“The raw technological horsepower is terrific, but it’s not going to determine how quickly AI transforms the economy,” said Andrew McAfee, a principal research scientist and co-director of the Massachusetts Institute of Technology’s Initiative on the Digital Economy.

Still, some businesses are finding ways to incorporate AI — although in most cases the technology is still a long way from replacing workers.

One company where AI’s promise and flaws are playing out is USAA, which provides insurance and banking services to members of the military and their families. After several pilot projects, some of which it closed down, the company introduced an AI assistant to help its 16,000 customer service workers provide correct answers to specific questions.

USAA is tracking its AI investments, but does not yet have a calculation of the financial payoff, if any, for the call center software. But the response from its workers, the company said, has been overwhelmingly positive. While it has software apps for answering customer questions online, its call centers field an average of 200,000 calls a day.

“Those are moments that matter,” said Ramnik Bajaj, the company’s chief data analytics and AI officer. “They want a human voice at the other end of the phone.”

Two years ago, JPMorgan Chase, the nation’s largest bank, blocked access to ChatGPT from its computers because of potential security risks. Only a few hundred data scientists and engineers were allowed to experiment with AI.

Today, about 200,000 of the bank’s employees have access to a general-purpose AI assistant — essentially a business chatbot — from their work computers for tasks such as retrieving data, answering business questions and writing reports. The assistant, tailored for JPMorgan’s use, taps into ChatGPT and other AI tools, while ensuring data security for confidential bank and customer information. Roughly half of the workers use it regularly and report spending up to four hours less a week on basic office tasks, the company said.

The bank’s wealth advisers are also employing a more specialized AI assistant, which uses bank, market and customer data to provide wealthy clients with investment research and advice. The bank says it retrieves information and helps advisers make investment recommendations nearly twice as fast as they could before, increasing sales.

Lori Beer, the global chief information officer at JPMorgan, oversees a worldwide technology staff of 60,000. Has she shut down AI projects? Probably hundreds in total, she said.

But many of the shelved prototypes, she said, developed concepts and code that were folded into other, continuing projects.

“We’re absolutely shutting things down,” Beer said. “We’re not afraid to shut things down. We don’t think it’s a bad thing. I think it’s a smart thing.”

McAfee, the MIT research scientist, agreed.

“It’s not surprising that early AI efforts are falling short,” said McAfee, who is a founder of Workhelix, an AI-consulting firm. “Innovation is a process of failing fairly regularly.”